import base64
from io import BytesIO

import firebase_admin
import gradio as gr
import numpy as np
from fastai.vision.all import *
import json
from firebase_admin import credentials, firestore
import pathlib

temp = pathlib.PosixPath
pathlib.PosixPath = pathlib.WindowsPath

cred = credentials.Certificate("firebase_key.json")
app = firebase_admin.initialize_app(cred)
db = firestore.client()

learn = load_learner("model.pkl")
names = json.load(open("./translations.json"))


def classify_id(image):
    buffered = BytesIO()
    image.save(buffered, format="JPEG")
    img_str = base64.b64encode(buffered.getvalue())

    pred, idx, probs = learn.predict(np.asarray(image))

    db.collection("preds").add(  # inilo db
        {
            "image": img_str,
            "prediction": pred.title(),
            "time_added": firestore.SERVER_TIMESTAMP,
        }
    )

    return [
        names[pred]["id"],
        f"./audios/id/" + names[pred]["id"] + ".mp3",
    ]


def classify_en(image):
    buffered = BytesIO()
    image.save(buffered, format="JPEG")
    img_str = base64.b64encode(buffered.getvalue())

    pred, idx, probs = learn.predict(np.asarray(image))

    db.collection("preds").add(  # inilo db
        {
            "image": img_str,
            "prediction": pred.title(),
            "time_added": firestore.SERVER_TIMESTAMP,
        }
    )

    return [
        # names[pred]["id"],
        names[pred]["en"],
        f"./audios/en/" + names[pred]["id"] + ".mp3",
    ]


with gr.Blocks(
    css=".gradio-container {background-image: url('file=Background/Fruitzone.jpg');background-size: cover; background-size: 100% 100%;}.block.svelte-kz0ejz{background-color: rgba(0,0,0,0);}"
) as demo:
    with gr.Row():
        with gr.Column():
            image_input = gr.Webcam(label="Gambar", shape=(200, 200), type="pil")
            predict_id_btn = gr.Button("Bahasa Indonesia", variant="primary")
            predict_en_btn = gr.Button("Bahasa Inggris", variant="secondary")

        with gr.Column():
            id_fruit_name = gr.Label(label="Bahasa indonesia", visible=False)
            id_audio = gr.Audio(label="Audio indonesia", visible=False)
            en_fruit_name = gr.Label(label="Bahasa inggris", visible=False)
            en_audio = gr.Audio(label="Audio inggris", visible=False)

    predict_id_btn.click(
        fn=classify_id,
        inputs=image_input,
        outputs=[id_fruit_name, id_audio],
        api_name="classify_image",
    )
    predict_en_btn.click(
        fn=classify_en,
        inputs=image_input,
        outputs=[en_fruit_name, en_audio],
        api_name="classify_image",
    )


demo.launch()